Microchannel Reat Sinks - MIT Lincoln Laboratory

Transcription

RICHARD J. PHILLIPS
Microchannel Reat Sinks
Microchannel heat sinks can be used in a wide variety of applications, including
microelectronics, diode laser arrays, and high-energy-laser mirrors. Heat sinks that
can be used to cool diode laser arrays have been fabricated in indium phosphide (lnP)
with a thermal resistance as low as 0.072°C/(W/cm2 }, which allows these devices to
dissipate loads in excess of 1,000 W/cm2 • This thermal resistance is nearly two orders
of magnitude lower than that achieved by the methods presentlyused in the microelectronics industry. A heat-:sink thermal- and fluid-performance model is presented;
microchannel fabrication techniques are described for loP and aluminum.
The contradictory requirements ofhigh heat
dissipation and low temperature rise are commonly demanded for a wide variety of applications. (For a description of some of the potential uses for microchannel heat sinks, see Box,
"Appli~tions.") High-energy-Iaser mirrors, surface-emitting laserdiode arrays, and high-speed
microelectronic devices, for example, can dissipate 10 to 1,000W/cm2 , yettemperature rises
generally must not exceed 100o e. The unique
- construction of microchannel heat sinks enables these elements to bring a heat source and
a heat sink into very close thermal proximity,
thus minimizing thermal resi§tance.
A microchannel heat sink consists of very
small channels and fins in a parallel arrangement (Fig. 1). The heat sink is very close to the
device that requires cooling; conduction transfers heat from the active device to the heat sink.
Heat is then transferred from the heat sink to
the outside world bypassing a coolant, usually
a liquid, through the channels.
The distinguishing feature of microchannel
heat sinks is the size oftheir cooling channels.
The same microfabrication techniques that are
used for the fabrication of electronic devices
can be used to make heat sinks 11-4]. These
techniques permit reliable fabrication of highprecision channels with channel widths, WC'
and fin widths, Ww ' as small as 50 ""m and
channel heights, b, as great as several hundred
microns. When liquids like water are forced
through these channels under pressures on
the order of 10 psi. heat transfer rates can rival
boiling heat transfer. (To withstand the coolant
The Lincoln Laboratory Journal, Volume I, Number 1 (1988)
pressure, the substrate must have a thickness,
t. on the order of 100 ""m).
CONVECTIVE FLUID FRICTION AND
BEAT TRANSFER
As with other heat-dissipation analyses,
thermal and fluid perlormance for microchannel heat sinks can be modeled with Reynolds
numbers, apparent friction factors, Nusselt
numbers. andparameters that describe the thermal and fluid perlormance of the coolant. The
thermal performance depends, in part, on whether the coolant flow is laminar or turbulent.
As coolant enters a channel (Fig. 2), a boundary layer develops within the coolant. This
boundary layer separates the uniform-velocity
core from the stationary coolant at the wall (the
"no-slip" boundary condition). The boundary
layers will increase in thickness as the coolant
flows through the channel. until the velocity
profile achieves a "fUlly developed" profile,
which does not change with downstream distance. The flow upstream from that point is
said to be "developing"; the flow downstream is
"fully developed."
The Reynolds number, Re, which is the ratio
ofinertial to viscous fluid forces. canbe used to
estimate the transition between laminar and
turbulent flow:
where V is the average coolant velocity in the
channel. De is the equivalent diameter of the
31
Phillips - Microchannel Heat Sinks
Applications
experience heating rates of 10 to
100 W/cm 2 • In spite of this high
thermal flux. the temperature variation of these mirrors must be
minimal. Distortion ofas little as
250 A can seriously degrade the
optical performance of a highenergy-laser mirror.
Microchannel heat sinks offer
solutions to thermal dissipation
problems for many existing and
developing applications. As
shown in the Fig., heat dissipation rates in computer modules
can reach levels that equal those
encountered by space vehicles
upon reentry into the earth's atmosphere [18]. But unlike reentry vehicles, the maximum surface temperature of the ICs in
computer modules usually must
be less than 100°C. In contrast,
reentry vehicles are designed to
withstand temperatures of more
than 1,OOO°C.
Because high-energy-Iaser mirrors typically absorb about 0.1 %
ofincidentlaserenergy, they can
Arrays of surface-emitting
GaInAsPIInP diode lasers are being developed as laser pumping
sources, optical interconnects
for ICs, and as two-dimensional
communication-switching devices. These arrays require highcapacity, high-efficiency heat
sinks.
computer modules. The highest
heat dissipation demonstrated
to date is 40 W/cm 2 for a singlechip module. Multichip modules
have somewhat lower heat dissipation figures. When averaged
across an entire multichip module, heat dissipation is generally
less than 3 W/cm2 • Use of microchannel heat sinks could accommodate more than 1,000 W/cm 2 •
Prototypes of microchannel
heat sinks for commercial applications are under development.
Devices may be marketed by
1990 - less than ten years after
the frrst development ofthe technology.
The Table lists thermal parameters for some state-of-the-art
q"
W/cm 2
Rocket Nozzle Throat (Internal)
.
.
.
103
10 2
•
Ballistic Entry
~
Reentry from Earth Orbit
/
. /.
Rocket Motor Case (External)
10- 1
Solar Heating
I
L..-.....L._....I---l_.....L._.L....--l_-'-_.....--I._-'
o
1000
2000
3000
4000
Temperature (K)
Very large scale integrated circuits dissipate on the
order of 10 W/cm 2 Within the next decade, some chips
will dissipate 100 W/cm 2 - as much heat as a nuclear
blast [18].
32
Technology
Maximum
Chip q"
(W)
Average
Module q"
(W)
R eoe
(Zc/W)
Nuclear Blast (1 Mt. 1 mil
/.
10
10- 2
Thermal Parameters for Computer Modulesa
Single Chip
Hp Finstrate
Hitachi 5-810
Burroughs PGA
Motorola MCA-2
Sperry Compact HX
Multichip Module
Mitsubishi HTCM
NEC SX LCM
Hitachi RAM
IBM 4381
IBM 3090 TCM
10.1
13.1
24.3
24.5
40
6.25
>8.4
13.1
17.0
29.8
<8.7
7.0
12
3.3
4.9
0.83
1.6
0.8
2.2
2.2
7.3
5
34.7
17.0
8.7
·See Ref. [t 8]
The Lincoln Laboratory Journal, Volume 1, Number 1 (1988)
IC Elements
Forming Surface
Heat Source
......1 --------- L - - - - - - - - - 1..... 1
1
/
...---------------------------,
Microchannel
Heat Sink
Cover Plate
Outlet
Flow
Inlet
Flow
IC Elements Forming
Surface Heat Source
Microchannel
Heat Sink
-.L \ W,--IIt
'Manifold
Block
T~~~~~~~~~
\
~
b
T
Cover Plate
Manifold Block
Fig. 1 - These views of a microchannel heat sink include the side view, which shows the longitudinal coolant flow
through a heat sink with channel length, L. The bottom, "end-on" view shows the channel width, We, fin width, W w ,
channel height, b, and substrate thickness, t.
IL..+..L-L.-~'--""=~L.-L.-"'-..L-..L-"""'=_:_'_""""-""""--"--"-'"""""o:::::f_--'--'--';....
. ..---Chan nel
Coolant
Flow ----.~
.....i - - -
Wa II
Central
Streamline
Developing . .
Fully Developed
Flow
....--I---.~
Flow
Fig. 2 - As coolant flows through a channel, the flow pattern changes from "developing" to "fully developed" flow.
The distinctive feature ofdeveloping flow is a change in the velocity profile of the coolant. The profile vectors change in
the region of developing flow, but remain constant in the fully developed region.
33
If
channel [here De =lIb Wc /(2Wc + 2b)), and vfis
the kinematic viscosity ofthe coolant. Channel
flow is usually considered to change from laminar to turbulent flow at Re = 2,300.
The thermal and fluid performance characteristics of a coolant depend on both the flow
regime and on whether the flow is fully developed. In general, the rate of heat transfer
increases as the coolant velocity increases.
But as the coolant velocity increases, the coolant pressure-drop also increases, and greater
pumping power is necessary.
Table 1 -
Approximate Values of Convection
Heat Transfer Coefficients [7]
Mode
tor (5) can be expressed as
The coolant experiences a pressure drop LlP
of
L
~P = 4fapp (De)
(p f
2
5-25
10-500
100-15,000
2,500-35,000
5,000-100,000
Free convection, air
Forced convection, air
Forced convection, water
Pool boiling, water
Flow boiling, water
f app = A (Re*)B
where
V2)
fk
A= 0.0929 +
where fapp is the apparent friction factor, L is
the channel length, Pf is the coolant density,
and ~ is a units constant. The low-Reynold'snumber, turbulent-flow apparent friction fac-
10 2
1.01612
L/De'
B
and Re* is based on the "laminar equivalent
diameter" of a rectangular channel [(2/3) +
(11/24 a) X (2 - 1/a)) De .InFig.3, the laminarflow apparent friction factor is plotted as a
function of channel aspect ratio (a = b/Wc )'
The rate of heat transfer per unit area, q",
from the channel walls to the coolant is proportional to the heat transfer coefficient, h:
q" = h
0:
a.
a.
a::::; 0.1
~
10.0
Nu= hDe
kf
a = 5.0
101
(1)
a = 2.0
where kcis the thermal conductivity ofthe coolant. The Nusselt number for turbulent flow is
a = 1.0
(6)
L...-..................................u....----'-......................."""---'--L..............LL..L.I.I
10- 3
10- 2
10- 1
Re/(L/De)
Fully Developed Flow
100
+
---.J
Fig. 3 - Laminar-flow apparent friction factor varies
with channel aspect ratio and Reynolds number. A channel aspect ratio of 1: 1 (channel width equals channel
height) provides the lowest fapp Re product.
34
~T
where LlT is the temperature difference between
the channel wall and the coolant. The heat
transfer coefficient is usually expressed in the
form ofthe dimensionless Nusselt number, Nu:
Q)
.....ro
__ -0 2680 _ 0.31930
.
L/De'
Nu = 0.012 [1.0 +
(~e) 2/3]
(ReO.87 _ 280) Pr>.4
(2)
where x is the distance from the channel entrance and Pr is the Prandtl number ofthe coolant (the ratio ofhydrodynamic to thermal diffusivities ofthe coolant). The Nusselt number for
laminar flow with several channel aspect ratios
is illustrated in Fig. 4.
As Fig. 4 shows, a microchannel heat sink
with an aspect ratio of 4 can have a fully developed Nusselt number of 5.35. Using Nu = 5.35,
kr = 0.6 W/m-oC (liquid water), and assuming a
channel equivalent diameter (De) of 100 p'm,
Eq. 1 gives a heat transfer coefficient of about
32,000 W/m2 _oC.
Table 1 compares the approximate range of
heat transfer coefficients commonly found in
other modes of heat transfer [7]. The rate of
heat transfer from microchannel heat sinks is
orders of magnitude higher than for forced-air
convection, and is comparable to that ofboiling
heat transfer. But because impulsive expansions are associated with boiling heat transfer,
this method can reduce the reliability of integrated circuits (ICs).
THERMAL RESISTANCE MODEL
A close look at the thermal model of a heat
sink reveals the key to the microchannel heat
sink's performance advantage. The thermal performance of a heat sink is often specified in
terms of its total thermal resistance. Several
common assumptions are used to construct a
thermal-resistance model. For example, it is
assumed that the coolant flow is steady and
incompressible, that the surface heating is spatially and temporally constant, that the temperature gradient through the fin thickness is
small compared with that along the fin height,
and that the fin tip is insulated (no heat transfer
from the cover plate). Conduction ofheat in the
streamwise direction in the fin and the coolant
is ignored. It is also assumed that the coolant
temperature and the heat transfer coefficient
are uniform over the entire channel surface at
any given distance from the channel entrance.
Based on a single channel and an adjacent
fin, the total thermal resistance, R" tot, is
R"tot=L(Ww +
wcl
a
)(
~
10.0
:J
Z
a = 3.0
a = 2.0
a = 1.0
3 '----'---l...J.....L..L.U-l.I...----l---'-..L-L..LJ..LJ.J....----L--'-....L..J...L..l..U.J
10- 3
10- 2
10- 1
100
L/(DeRePr)
Fig. 4 - Deeper channels increase the Nusselt number.
~Tsurf
= ~
where L:1 T surf is the peak surface-temperature
rise above the inlet coolant temperature (ignoring viscous dissipation heating), Q is the total
heating rate over the channel and fin pair, and
q" is the uniform surface heating rate per unit
area
The total thermal resistance can be divided
into several components. Figure 5 identifies six
of these components. If many discrete heat
sources are used, the thermal-spreading (heat
dispersion) resistance, R" spread' distributes the
heat flow until the heat flow just below the surface of a discrete heat source (such as an IC) is
uniform. This thermal-spreading resistance is
R" spread-
10
~TSurf)
(Q
L(Ww + wcl
c kw a
where c is a constant that depends on the
shape ofeach heat source (eg, c =4 for a square
heat source [8]), kw is the thermal conductivity
of the material, and a is the characteristic
dimension of the heat source shape.
Conduction through the solid material between the heat-source surface and the channel-base-fin-base plane adds another component to the thermal resistance. This resistance,
R" solid' is
R"solid = (
~)l + (~)2
where t is the thickness ofeach solid layer, and
kw is the thermal conductivity of the material.
35
IC Chip
Thermal
Interface
Microchannel
Heat Sink
•
Coolant Flow
Fig. 5 -
Thermal resistance components.
The interlace between the heat source and
the heat sink gives rise to the interlace thennal
resistance. R" into A microchannel heat sink
can be manufactured as an integral part of an
Ie. which eliminates this component of thermal resistance. However. if a heat sink is built
separately and then attached to a heat source (a
"cold plate" arrangement) then the value of
R" int depends upon the attachment method.
Various thennal interlaces, such as solder and
epoxy bonding. thennal grease, and gas layers
are in common use. A newly developed method.
the microcapillary thennal interlace. provides
very low thennal resistance [4.9]. (For more
details. see Box. "Thermal Interlace Techniques.")
Fins transfer more heat than the fm-base surface could if it were directly exposed to the
coolant. The heat flow is therefore funneled
36
into the base of the fin. The thennal resistance
due to constriction. R"con. is [10]
R"
con
=
(Ww + we!
1Tkw
1
In - - - - - - - sin [1TWw/(2Ww + 2We!l
This equation assumes that all of the heat is
conducted through the fin before being convected to the coolant. In reality. some heat is
convected from the channel base, thereby making the constriction thennal resistance somewhat smaller than predicted. This difference is
small and is ignored for simplicity.
A convection thennal resistance. R" conv' is
associated with heat conduction through the
fin and convection from the fin and channel
base surlaces. The value of this resistance
component is
R"conv=
(Ww + we!
h Wc + 2 h b 17f
(3)
Thermal Interface Techniques
The thermal resistance between a heat source and a heat
sink can be lowered in several
ways. Bonding methods (epoxy,
solder, etc.) form a solid interface. A solid interface, however,
is subject to mechanical stress
due to the differences between
the coefficients ofthermal expansion of a heat sink and a heat
source. And a solidbond can have
gas voids, which act as thermal
insulators. Furthermore, a solid
bond can be formed only once a heat sink cannot be easily removedwithout damaging itselfor
the heat source.
Gas-layer interfaces, such as
He, are difficult to implement because they require a microscopic
interface gap between optically
flat dust-free surfaces. Liquidlayer interfaces, such as thermal
grease, often require large compressive loads to keep the interface gap small.
The microcapillary interface,
developed at Stanford University,
provides a low-resistance interface that has none of the above
problems [4,9). As shown in the
Fig., this interface uses reentrant channels that are partially
where
l1f
filled with a high-surface-tension
liquid. The surface tension develops a strong attractive (suction) force between the two surfaces, resulting in an interfacial
gap that is typically less than 0.7
J.Lm. The interface is reusable; the
heat sink and the heat source
can be separated without damage
to either device. Because the in-
terface is a liquid, differences in
thermal coefficients of expansiondo notcause problems. Moreover, the close spacingofthe heat
source and heat sink does not
require optically flat surfaces or
large compressive loads. Thermal resistance as low as O.02°C/
(W/cm2 ) have been measured for
this type of interface.
Reentrant
Capillary
Channels
I
To Microchannel
Heat Sink
•
This microcapillary interface features a thin gap with
low thermal resistance.
is the fin efficiency:
_ tanh (mb)
_ ( 2h ) 0.5
mb
,m - kwWw
T/f-
The first term in the denominator of Eq. 3
represents convection from the channel base;
the second term represents convection from
the fin.
The final thermal resistance component is
associated with the bulk temperature rise of a
liquid coolant from the channel entrance. This
temperature rise is caused by absorption of
heat from fin and channel base surfaces. The
coolant bulk temperature rise thermal resistance, R"bulk' is
x(Ww + wei
R"bu1k= - - - - PfCpfbWc v
where Cpf is the coolant specific heat.
One phenomenon causes a temperature rise
independent ofthe heating rate, so it cannot be
formulated as a thermal resistance. This is the
coolant temperature rise due to the conversion
of mechanical energy (fluid pressure) into thermal energy (fluid temperature rise), ~Tvisc:
AP
ATvisc
= Pf Cpf J
37
Phi1Iips - Microchannel Heat Sinks
where J is the mechanical equivalent of heat
(1.0 N-m/J).
Combining the total thermal resistance components, the total temperature rise, LiTtot ' of
the heat source surface, including viscous dissipation effects is
LiTtot = R" tot q" +
shown in Fig. 7. The dotted curve of Fig. 6
shows the coolant temperature; the dashed
curve shows the surface temperature.
PERFORMANCE PREDICTIONS
Based on the thermal-resistance model summarized in Eq. 4, a computer program was written. This program predicts the thermal and
fluid performance of microchannel heat sinks.
A complete program listing appears in Ref. 5.
The program has been used to compute the
thermal and fluid performance for a wide variety of coolant flow-rate constraints and heatsink designs. It permits variable property effects in both the coolant and in the heat-sink
material.
The coolant is assumed to be a liquid, so
compressibility effects in the coolant are not
included in the computations. Furthermore,
R" spread and R" int are omitted from the calculations. R" spread usually is small and varies considerably with geometry; R" int varies considerably with the type of interface used.
(4)
Il.Tvisc
where
R" tot = R" spread + R" solid + R" int + R" con
+ R" cony
+ R"bulk
Figure 6 shows a typical surface temperature
distribution in the streamwise direction of the
coolant. The surface temperature increases
with distance from the channel entrance because the coolant temperature rises and because the heat transfer coefficient decreases
with distance from the channel entrance. (Nu
decreases with increasing x - see Eq. 2 and
Fig. 3.) If the microchannel heat sink is longer
than the surface heat source, some heat is
conducted laterally from the heat source [5), as
Predicted Peak Surface Temperature
~Extent of Heat source--.l
Predicted Surface
Temperature Profile
(Without Thermal
Spreading)
I
I
I
I
I
Actual Peak Surface Temperature
Actual Surface Temperature Profile
(With Thermal Spreading)
I
I
I
I
I
I
Tf,in
----t~~
•
-----1~~
Flow
Direction
Fig. 6 - Various thermal resistance components, the greatest being R"bulk and R"conv,
contribute to the temperature variations in the stream wise direction within a heat sink.
38
t
/
ttttt
Coolant Flow
----1.~
___-1:~
Flow
•
Direction ----1:~
a) Streamwise
Direction
b) Transverse
Direction
- - - - Temperature Profiles without
Thermal Spreading
Temperature Profiles with
Thermal Spreading
Fig. 7 - Effect of thermal spreading at the heat-source
perimeter.
Figures 8 and 9 show thennal resistance and
pumping power per unit area as functions of
channel width. (See Refs. 5 and 11 for additional performance prediction comparisons.)
The results for a reference-case heat-sink design are given by the three solid CUlVes in Fig. 8.
(R"solid and R"con are not shown because they
are small and vary little with channel width.)
These CUlVes show that R"bulk decreases with
increasing channel width because more coolant can be forced through a channel with a
larger equivalent diameter, for a given overall
pressure drop. R"conv increases for laminar
flow; it is somewhat smaller for turbulent flow.
Total thennal resistance is dominated by R"bulk
for small channel widths and by R"conv for
moderate to large channel widths. The results
show that a microchannel heat sink can have
an R"tot smaller than 0.1°C/(W/cm2 ). The required pumping power per unit surface area is
less than 10 W/cm 2 •
The blue and red CUlVes in Fig. 8 show the
effect on the heat-sink performance of changing the channel length. The thermal performance of the shorter channel design is superior, but this performance is obtained at the
expense of a large increase in the pumpingpower requirement. The thermal performance
of the longer channel design is not quite as
good, but the coolant distribution system to
and from the heat sink is considerably simpler.
If a heat sink is made of copper, the total
thermal resistance decreases - the thermal
conductivity of copper is 2.7 times that of silicon (red CUlVes in Fig. 9). To improve heat-sink
performance, a heat sink and an IC can be fabricated independently and then bonded via a lowthermal-resistance interface (such as a microcapillary interface). The copper microchannel
heat sink can provide better thermal performance than a heat sink directlyfabricated within the IC itself.
The orange CUlVes in Fig. 9 show the effect of
using Freon (CCI 2 F2 ) as the liquid coolant,
rather than water. Because of its low reactivity,
Freon is commonly used in direct-contact pool
boiling cooling for electronics. The thermal
performance of Freon is not quite as good as
the thermal performance of water, but the
pumping power requirements are similar.
The blue CUlVes in Fig. 9 show the result for
constant pumping power per unit heater-surface area, P"; the green CUlVes show the result
for constant coolant volumetric flow rate per
unit of heater surface, V". For IlXed pumping
power, the optimal laminar flow design occurs
at We = 50 ~m. For the fixed flow-rate constraint, the thennal resistance increases and
the pumping power decreases with channel
width. As expected, the constraints predict identical performance for P" = liP . V".
FABRICATION
Large arrays ofsurface-emitting GaInAsP IInP
laser diodes [12] can dissipate several hundred
WI cm2 • To address this coolingchallenge, microchannel heat sinks were fabricated in InP, the
39
100
1.0
Fully Developed
Developing
Laminar Flow ~ Laminar Flow
~
N
"-
E
u
........
,
" .....
S
u
~
~
u
c:
co
....
l/)
---..
Laminar Turbulent
Flow
Flow
Q)
l/)
-- --- --
u
~
Q)
l/)
-- ... _-
c:
Q)
I-
....co
::::l
I-
Q)
R"tot
0.1
10
Q)
c:
E
I-
R"conv
co
E
Q)
I-
The Following Designs Are Identical to
the Reference Case*Except as Indicated
I-
....co0
0.01
100
0
100
N
E
u
..
........
10
I-
Q)
..
......
..
200
400
300
Channel Width (,urn)
500
::::l
(/)
~
co
Q)
..•........• L =0.1 cm
----L = 10.0 cm
f-
Q)
u
co
................... ,...
l-
~
S
a.
Q)
0-
1.0
600
.. ...
.-
~
0
0C)
c:
a.
1.0
-- -- -- ---
E
::::l
0-
0.1
0
100
200
300
400
Channel Width (,urn)
500
600
- - - * Reference Case: Water-Cooled Silicon
Tf in =300K, liP =68.9 kPa (10 psi), q" = 100 W/cm 2,
L = 1.0 cm, a =4.0, Ww/W c = 1.0, t = 100 ,urn,
Fully Developed Turbulent Nusselt Number,
Kc =Ke =KgO =0
Fig. 8 - Total thermal resistance and pumping power requirements are shown as functions of channel width. The
pressure drop is held constant for these curves.
40
1.0 r - - - - - - - - - - - - - - - - - - - - , 100
Fully Developed~~~ Developing
Laminar Flow
Laminar Flow
I
./ Turbulent Flow
\, \ ' ( "
.
Results
\
co
en
en
Q)
a:
:
::::J
...~
.....~_---=-_===_
'-
~-..
/laminar Turbulent
co
...
....co
...
Q)
'-:'. . I . . . . . . . . .----------'""'1
...
'.
0.1
a:
Results
6,
c:
en
~ Laminar Flow
"
Q)
()
Q)
..
...E
----
Q)
Co
10
The Following Designs Are Identical to
the Reference Case*Except as Indicated
.............. V" = 10.53 (cm 3/s)/cm 2
- - - - - - P" = 0.726 W/cm 2
_ . _ . - Heat Sink Material is Cu
- - - - Coolant is Freon
Q)
£:
~
....co
o
~
0.01
100
L..-_--JL--._--J
o
100
E
Q)
~
Q)
()
........co
::::J
en
.::.t.
co
Q)
a..
1.0
600
--JL--._ _L - - _ - - J L - - . _ - - - '
200
300
400
Channel Width (J-Lm)
500
~-----------------~
N
E
()
........
S
...
10
Q)
5:
o
a..
Cl
c:
-~-tE~·~-----------------------".
.
Co
~
E
::::J
a..
~
\"
1.0
0.1
o
100
200
300
400
500
600
Channel Width (J-Lm)
----*Reference Case: Water-Cooled Silicon
Tf in =300K, ~P =68.9 kPa (10 psi), q" =100 W/cm 2,
L =1.0 cm, a =4.0, Ww/W c =1.0, t =100 J-Lm,
Fully Developed Turbulent Nusselt Number,
Kc =Ke =KgO =0
Fig. 9 - Total thermal resistance and pumping power requirements are shown for various coolants, heat-sink
materials, and coolant flow-rate constraints.
41
substrate material of these laser diodes. Other
researchers have produced microchannels in
Si, but this is the first demonstration of microchannel heat sinks in InP. A method was also
developed to fabricate microchannels in
aluminum.
Two methods have been used to fabricate
microchannels: precision sawing and orientation-dependent etching. Precision sawing
with semiconductor dicing saws is an efficient
production technique for Si. However, because
of its brittleness, sawing does not work well
with InP [5]; an InP heat sink's fins frequently
chip and break.
Unlike sawing, orientation-dependent etching works well in forming microchannels in InP
[5] and in Si [1,2,4]. Microchannels were etched
along the <011> direction in InP, using a 3:1
mixture (by volume) of H3 P0 4 :HCI at room
temperature. The channels were patterned in a
3,000-A layer of phosphosilicate glass (PSG),
applied to a polished chip surface that was preetched with full-strength potassium ferricyanide. (Pre-etch is required for good adhesion of
the PSG. Without the pre-etch, undesirable
undercutting occurs.) The channel depth etching rate averaged 0.62 j.lm/min, and the undercutting etching rate averaged 0.22 j.lm/min.
Figure 10 shows three views of InP microchannel heat sinks. The top view illustrates
that the channels do not extend to the edge of
the chip, which eliminates the need to seal the
channel ends. The channel-end shape is shown
in a streamwise view; the angle of the solid
material is 35° to 38°. This angIe helps deflect
the inlet and outlet coolant flow and therefore
helps reduce the pressure drop.
The wavy appearance ofthe fin tips in the top
view is due to drooping ofthe PSG mask (caused
byundercutting). This waviness produces roughness features in the channel walls, which propagate to the fin base as the 35° to 38° triangular
shapes seen in the streamwiseview. This roughness can serve as a turbulence promoter. Increased turbulence increases heat transfer
rates at the expense of increasing pumping
power requirements. The channel bottoms in
the top view have a "granular" texture. This
42
texture is due to small mounds created by
defects. These defects are in the polished surface before etching starts. The mounds also
have 35° to 38° edges.
The transverse view in Fig. 10 shows that the
channel and fin cross-sections are quite uniform. When the PSG mask adheres properly,
the channel uniformity is generally within 5%.
Channel uniformity of 5% or better is recommended because thermal performance (for the
same pumping power) decreases exponentially
with increasing nonuniformity [13].
As an alternative to forming microchannels
in InP chips, fabrication techniques were developed for cold plate microchannel heat sinks. A
numerical-control milling machine was used
to cut microchannels in aluminum [5]. An arbor
was constructed with five 250-j.lm (thickness)
X 3.175-cm (diameter) circular jewelers' saws,
each separated by 240-j.lm shim-stock. A milling rate of 1.25 cm/min produced heat sinks at
a rate of approximately 20 cm2 /h.
Increasing the number of saws will provide
the production rates necessary for mass production and commercialization. Other fabrication methods, including extrusion and laser
machining, may also be suitable for mass
production.
EXPERIMENTS
Thermal and fluid performance tests have
been conducted on microchannel heat sinks
fabricated directly within InP substrates. The
three-sided etched-channel chips were packaged to provide coolant flow to and from the
chip, as shown in Fig. 11. A cover plate with
inlet and outlet manifold slots was epoxied to
the chip. The cover plate is Corning #7059
glass, which has a coefficient of thermal expansion that closely matches the coefficient of
InP. Trabond F-113 epoxy, which wets InP surfaces well (ensuring strong bonds), was used to
bond the chip to the cover plate. The epoxy has
good stability at the chip operating temperatures for periods in the tens of hours. (No longterm tests have been conducted to date.) Devcon five-minute epoxy was used to attach the
cover plate to the Lexan coolant-manifold
block.
The Lincoln Laboratory Journal, Volume I, Number 1 (l988)
a)
Top View
(~21 O-J,Lm-Wide
b)
c)
Transverse View
Channels on 375-J,Lm Centers).
Streamwise View (b
(~160-J,Lm-Wide
~
160-J,Lm).
Channels on 250-J,Lm Centers, a - 1.0).
Fig. 10- Orientation-dependent etching produced these InP microchannels.
The resistor heat-source design that provided
the thennalload for the heat-sink tests is shown
in Fig. 12. Four 0.25 X 0.25-cm square resistors
were fabricated directly into the top surface of
the test chips. The resistance was typically 16
to 19 fl/square; the resistance uniformity
among the four resistors was betterthan ± 1.1 %.
Each resistor received power independently of
the other resistors, which allowed investigation of thermal spreading at each heat-source
perimeter.
The fluid performance of a typical microchannel heat sink is shown in Fig. 13. In this
figure, the pressure drop is plotted as a function of Reynolds number. The uncertainty in
the Reynolds number is primarily due to uncerThe Lincoln Laboratory Journal, Volume I, Number 1 (1988)
tainty in the channel equivalent diameter. The
uncertainty in the pressure drop is approximately equal to the size of the data points. The
predicted pressure drop agrees, within 5% to
10%, with the experimental data for laminar
flow. (Roughness does not seriously affect the
apparent friction factor in laminar flow.) The
turbulent flow pressure drop predictions, however, are low because those predictions assumed that the channel surfaces were smooth.
The apparent friction factor is larger than it is
for smooth surfaces because the roughness
features extend past the turbulent-flow laminar sublayer.
The thermal performance of a microchannel
heat sink fabricated in Si is shown in Fig. 14.
43
Microchannel
Heat Sink
Cover Plate --".~f---""'"""'I'~rr-~
Return
- - - - , / 1 - - - - Manifold
Pressure Tap
Manifold
Substage
#2
~
Supply
- / - - - - - - ---Manifold
...
I ------~--T"""_--_f'
Pressure Tap
I
I
I
I
I
I
I
Manifold
Substage #1
Coolant
Return
Coolant
Supply
--..
I
I
I
I
I
I
I
I
1
Manifold
~
Substage #0
Fig, 11 -
This coolant-distribution manifold supplies coolant to microchannel heat sinks during thermal tests,
The thennal performance predicted using the
theory in this article is presented for comparison. Because the heat source is large, thermal
spreading does not seriously affect the thennal
performance.
InP microchannel heat sinks were tested at
heat dissipation rates as high as 1,056 W/cm2
with thermal resistance as low as 0.072°C/
(W/cm2 ). A contour map of typical surface temperature profiles for microchannel-cooled InP
chips is shown in Fig. 15a This figure shows
44
how thermal spreading affects the thermal
performance of a heat sink. Thermal spreading influences thermal performance because,
in this case, the thennal-spreading diffusion
length is comparable to the size of the resistor
heat sources. Figure 15b compares theory and
experiment - the predicted total thennal resistance is approximately 20% larger than the experimental result. This discrepancy is thought
to be the result ofthe use oftwo superimposed,
one-dimensional thermal-spreading models to
•
a)
Photographs of a Typical Test Chip
25
~O.25 cm-.j
J.lm AI Wire
/ I T O Gold-Plated Alumina!
Gold Metallization
~ f.-1.0J.lm
S.+---Semi-Insulating InP
Resistor Definition Slots
View A-A (Enlarged)
Flow Direction
for Transverse
Thermal Spreading
~
Tests
O.25-cm Square Resistor (Typ.)
Flow Direction
for Streamwise
Thermal Spreading--..
Tests
A
Gold Metallization (Typ.)
A ~tftl""'I--+---- 25 J.lm AI Wires (Typ.)
(Approx. 15 Wires/mm of
Metallization Length)
~ Gold-Plated
.....1------- Power
Alumina (Typ.)
Supply Wire (Typ.)
, ' - - - - - - - Metallization Voltage Wire (Typ.)
b)
Schematics
Fig. 12 - These test-chip resistors provided thermal loads to heat sinks to measure the heat sinks' dissipation
capability.
45
80
Ul
c.
c.
60
0
....
0
OJ
....
40
Coolant = Water 20°C
W w = 155 J.Lm
We = 220 J.Lm
b = 165 J.Lm
L = 0.97 cm
Table 2 Experimental Performance Parameters
:::::l
Ul
Ul
OJ
....
a..
Ref.
Heater Size
(em X em)
Ww
(pm)
We
(pm)
(pm)
L
(em)
[14]
[15]
0.508
0.7
127
2,540
127
5,870
12.700
1,000
0.635
5.0
[16]
0.8
2,540
800
400
b
20
(est.)
8.3
(est.)
[17]
0
0
1000
2000
3000
Reynolds Number
Fig. 73 - Predicted and experimental values of pressure drop are plotted as functions of Reynolds number.
N
E
u
~ 0.3
::::::-
u
OJ
u
0.2
c
....co
We = 64 J.Lm
W w = 136 J.Lm
b = 245 J.Lm
t = 244 J.Lm
L = 1.0 cm
Ul
~ 0.1
0::
co
E
....
OJ
~
~
....oco
0.8
1.0
340
35
340
55
900
400
[5]
[19]
0.25
3.8
155
100
220
200
165
1,700
Ref.
Coo/ant
Pressure
Drop
(kPa)
0.002
0.006
0.010
Distance from Upstream Heater Edge (m)
~
- - - Theoretical Values
---Experimental Data Points Obtained
from Ref. 4
V" = 1.277 (cm 3 /s)/cm 2
q" = 34 W/cm 2
2.4
2.00
(overall)
[14]
[15]
[16]
[17]
[4]
[5]
[19]
~
2.....
[4]a
Air
Water
Water
Water
Water
Water
Water
Air
1.2
110
20
20
365
267
0.97
5.0
Therma/
Flow
Rate
Resistance
(em 3/sj/em 2 [OC/(W/em 2)]
1,940
7.9
0.21
4.73
11.3
4.4
88.9
088
0.17
2.1
0.32
0.083
0.072
0.29
10.1
a35-l'm (length) interrupted cooling fins as opposed to continuous fins.
because it depends on the applications for
which the heat sinks were designed. For example, some heat sinks were designed as multichip modules [15-17], which do not perform
quite as well as single-chip modules [4,5]. Part
of the performance advantage of single-chip
modules comes from fabricating microchannels directly within a chip. The channels of
multichip modules, in contrast, are fabricated
in a cold plate to which the chips are attached.
CONCLUSIONS
Fig. 74 - The local thermal resistance of a watercooled Si microchannel heat sink increases with increasing distance from the upstream heater edge.
predict lateral thermal spreading at the heatsource perimeter, instead of a detailed twodimensional model [5].
Table 2 lists experimental results reported
by several researchers who have studied microchannel heat sinks. The relative performance
of the heat-sink designs is difficult to rank,
46
Microchannel heat sinks dissipate large
amounts of heat with relatively little surface
temperature rise. These heat sinks are useful
for a wide variety of applications, including
microelectronics, diode laser arrays, and highenergy-laser mirrors. The thermal performance
of microchannel heat sinks is approximately
two orders of magnitude better than the thermal performance of other devices currently
used to cool microelectronic devices. Furthermore, the pumping power required to force
tI
Flow
Direction
q" = 170.9 W/cm 2
a)
Surface Temperature Contours for
One Heater Energized
W c = 220,um t = 255 ,um
W w = 155,um L = 0.97 cm (Overall)
b = 165 ,um
= 0.25 cm (Heated)
0.25.--------------...,
Prediction
0.20Including
Thermal
/
~ 0.15~ Spreading
c
....coCIl
CIl
f
0.10 -
Q)
a:::
co
E
...
Q)
0.05-
o
I
In
I
..
I
I
I
I
I
I
I
0.5
1.5
2.5
3.5
I
..c
I-
-1.5
~
Distance from Upstream Heater Edge (mm)
o
-0.5
4.5
I-
b)
Comparison Between Data and Theory [5]
Fig. 15 - a) The effects of thermal spreading are shown
by the contours in this photograph. One heater is operating; the other three are off. b) Predicted and experimental values of total thermal resistance are plotted as functions of distance from the upstream heater edge.
liquid coolants through these heat sinks can
be kept to less than 10 W/cm2 •
By using orientation-dependent acid etching,
microchannel heat sinks can be fabricated in
<100> InP. The thermal performance of InP
microchannel heat sinks is excellent - thermal resistance as low as 0.072°C/(W/cm2 ) and
heat dissipation as high as 1,056 W/cm 2 have
been measured for InP heat sinks. Fabrication
techniques have also been developed for mass
production of microchannel heat sinks in aluminum and other metals.
REFERENCES
1. D.B. Tuckerman and RF.W. Pease, "High-Performance
Heat Sinking for VLSI," IEEE Electron Device Lett.
EDL-2. 126 (1981).
2. D.B. Tuckennan andRF.W. Pease, "Ultrahigh Thermal
Conductance Microstructures for Cooling Integrated
Circuits," Proc. 32nd Electronics Components Corif-,
p. 145 (1981).
3. D.B. Tuckerman and RF.W. Pease, "Optimized Convective Cooling Using Micromachined Structures," Electrochemical Soc. Extended Abstracts 82.197 (1982).
4. D. B. Tuckerman. "Heat-Transfer Microstructures for
Integrated Circuits," PhD Thesis, Stanford University,
Stanford, CA (1984).
5. RJ. Phillips, "Forced-Convection. Liquid-Cooled, Microchannel Heat Sinks," Masters Thesis, Massachusetts
Institute of Technology, Cambridge, MA (1987).
6. V. Gnielinski, "New Equations for Heat and Mass Transfer in Turbulent Pipe and Channel Flow," Int. Chem.
Eng. 84. 82 (1962).
7. J.P. Holman, Heat Transjer, McGraw-Hill, New York
(1976).
8. R C, Joy and E.S. Schlig, "Thermal Properties of Very
Fast Transistors," IEEE Trans. Electron Devices ED17.586 (1970).
9. D.B. Tuckerman and RF.W. Pease, "Microcapillary
Thennal Interface Technology for VLSI Packaging."
Symp. on VLSI Technology, Digest, p. 60 (1983).
10. A.D Kraus and A. Bar-Cohen, "Thermal Analysis and
Control of Electronic Equipment," McGraw-Hill, New
York (1983).
U. RJ. Phillips, L.R Glickman, and R Larson, "ForcedConvection, Liquid-Cooled, Microchannel Heat Sinks
for High-Power-Density Microelectronics," Proc. ojthe
Int. Symp. on Cooling Technology jor Electronic
Equipment, Pacific Institute for Thermal Engineering,
Honolulu, HI (1987),
12. Z.L. Liau and J.N. Walpole, "Surface Emitting GaInAsPflnP Laser With Low Threshold Current and High
Efficiency," Appl. Phys. Lett. 46. U5 (1985).
47
Symbols
a
b
Cpf
De
fapp
&
h
J
kf
kw
L
m
Nu
P"
Pr
q"
Q
R"bulk
R"eon
R"eonv
Heat source characteristic
dimension [m]
Channel height [m]
Coolant specific heat [J/kg_0C]
Channel equivalent diameter [m]
Apparent friction factor
Units constant [kg-m/N-s 1
Convective heat transfer coefficient
[W/m2 _0C]
Mechanical equivalent of heat
[N-m/J]
Coolant thermal conductivity
[W/m-°C]
Heat-sink material thermal conductivity [W/m-°C]
Channel length [m]
Fin efficiency parameter [m- l ]
Nusselt number
Pumping power per unit surface area
[W/cm2 ]
Coolant Prandtl number
Heat transfer rate per unit area
[W/m2 j
Total heating rate [WI
Coolant bulk temperature rise thermal resistance [OC/(W/cm2 )]
Contraction thermal resistance
[OC/(W/cm2 )]
Convective thermal resistance
[OC/(W/cm2 )]
13. R.K. Shah and A.L. London, "Effects of Nonuniform
Passages on Compact Heat Exchanger Performance,"
J. Eng. Power 102, 653 (1980).
14. N. Goldberg, "Narrow Channel Forced Air Heat Sink,"
IEEE Trans. Components Hybrids Manuj. Techno!.
Interface thermal resistance
[OC/(W/cm2 )]
Solid
material thermal resistance
R"solid
[OC/(W/cm2 )]
R"spread Spreading thermal resistance
[OC/(W/cm2 )]
Total
thermal resistance
R"tot
[OC/(W/cm2 )]
Re
Reynolds number
Solid material thickness [m]
t
V
Average coolant velocity [m/s]
Coolant volumetric flow rate per unit
V"
surface area [(cm3/s)/cm2 ]
Channel
width [m]
We
Fin width [m]
Ww
x
Distance from channel entrance [m]
Fin efficiency
Tlf
Channel aspect ratio
Ct
6.P
Coolant pressure drop [N/m2 ]
Wall-to-coolant temperature differ6.T
ence [0C]
Peak surface temperature rise above
6.Tsurf
inlet coolant temperature [0C]
Total temperature rise [0C]
6.Ttot
Viscous heating coolant temperature
6.Tvisc
rise [0C]
Coolant kinematic viscosity [m2 /s]
Vf
Coolant density [kglm3 ]
Pf
R"int
and in developing bonding techniques and P. Doherty
and D. Dufour for their help in developing the machining process for mass production of microchannels in aluminum.
CHMT·7, 154 (1984).
15. D. Nayak, L.T. Hwang, I. Turlik, and A. Reisman, "A
High-Performance Thermal Module for Computer Packaging," J. Electron. Mater. 16, 357 (1987).
16. T. Kishimoto and T. Ohsaki, ''vLSI Packaging Technique Using Liquid-Cooled Channels," Proc. 1986 IEEE
Electronic Component Colif., p. 595 (1986).
17. S. Sasaki and T. Kishimoto, "Optimal Structure for
Microgrooved Cooling Fin for High-Power LSI Devices," Electron Lett. 22, 1332 (1986).
18. A Bar-Cohen, "Thermal Management ofAir- and LiquidCooled Multi-chip Modules," IEEE Trans. Components
Hybrids Manuj. Technol. CHMT-IO, 159 (1987).
19. M. Mahalingam, "Thermal Management in Semiconductor Device Packaging," Proc. IEEE 73, 1396 (1985).
ACKNOWLEDGMENTS
I wish to thank Dr. J Walpole for his help in developing channel fabrication in InP. S. Connor for fabricating the test-chip resistors. and P. Daniels for his
help with the packaging design. I also thank D. Calawa and S. Hoyt for help in sawing channels in InP
48
RICHARD J. PHILLIPS has
worked for Lincoln Laboratory since 1983. A member
of the Optical Systems Engineering Group, he studies
beam-path conditioning systems for high-energy laser
systems. Rick received a
bachelor's degree in mechanical engineering from Colorado State University in 1982 and a master's degree in
mechanical engineering from MIT in 1987. His varied leisure time activities include canoeing, fishing, scuba diving, and travel.
The Lincoln Laboratory Journal. Volume 1. Number 1 (19881
I
•

Microchannel Reat Sinks - MIT Lincoln Laboratory

Transcription

Similar documents

EXTENDED LIFE ORANGE

Hangsterfer`s S787R New Technology Coolant

price list

VAG Coolant Application Guide G12 G12++ G13

November 2012 - Harry Warren, Inc.

EMMEDUE®: Green Culture for a Future to build.

Supporting information for Carbon fiber solder matrix composite for

RC3017

- My Wellness Center

Issue 12

THERMOPHYSICAL PROPERTIES OF TROPICAL FRUIT JUICESl